Claude Code Desktop

Two desktop clients, two purposes

The term "claude desktop" covers two distinct products, and getting them mixed up is easy because they share a name root and both run as native applications.

Claude Desktop is the general-purpose chat application. It targets everyday conversation, document review, and Q&A workflows where you want a windowed interface rather than a browser tab. It connects to the same Claude AI models as the web client and supports MCP server connections through a config file, which lets it reach external tools and data sources. This is the client most non-developer users mean when they say "claude desktop."

Claude Code Desktop is the windowed companion to the Claude Code CLI. It adds a graphical context browser, inline diff view, project settings panel, and session history to the same agent-based coding workflow you get from the terminal. Under the hood it bundles a copy of the CLI runtime, so you get the full Claude Code functionality without opening a terminal — though most developers run both in parallel.

Both clients are available as downloadable installers. The claude ai download page lists the current artefacts with file formats and checksum styles.

What Claude Code Desktop adds over the CLI

The CLI is the primary interface for Claude Code and the one most guides assume. The desktop client adds convenience features on top without replacing anything.

The graphical context browser lets you see which files and symbols the model has read in the current session, highlighted in a file tree. On a large project where the model has pulled in dozens of files, this panel makes it clear what context is live without reading through the terminal scroll buffer.

The inline diff view renders proposed file changes as a side-by-side diff inside the client window, with one-click accept or reject per change. In the terminal the same changes are printed as text output; the desktop makes reviewing them faster for engineers who prefer visual diffs.

The project settings panel exposes the same configuration that lives in the CLI's config file — model selection, context window budget, skill paths — through a form interface. Changes write back to the same config file, so they are visible from the CLI too.

For teams, the desktop client also shows the session history from any shared workspace, which is useful during code review when you want to trace what the model did during a previous session without re-running it.

Platform and version overview

The table covers platform availability, client type, and the version format used across installers and the npm package.

Platform	Client type	Version format
macOS (Apple Silicon)	Claude Code Desktop (.dmg), Claude Desktop (.dmg)	MAJOR.MINOR.PATCH (e.g. 1.5.0)
macOS (Intel)	Claude Code Desktop (.dmg), Claude Desktop (.dmg)	MAJOR.MINOR.PATCH — same cadence as npm
Windows 10/11 (x64)	Claude Code Desktop (.exe), Claude Desktop (.exe)	MAJOR.MINOR.PATCH — installer and npm match
Linux	CLI only (no desktop installer)	npm MAJOR.MINOR.PATCH via npm package

Version numbering across the product family

Version numbers follow a consistent pattern across the desktop installer, the npm package, and the CLI binary. When Anthropic ships version 1.6.0, the desktop DMG, the Windows EXE, and the npm package all carry the same 1.6.0 tag. This means you can match the desktop About panel to the claude --version output and know immediately whether the two are in sync.

The desktop app checks for updates automatically at launch and prompts you when a new version is available. If you prefer to control updates manually — common in corporate environments with an approval workflow — you can disable auto-update in the app settings and pull new versions from the claude code download page directly.

MCP server support in Claude Desktop

Claude Desktop can connect to Model Context Protocol servers, which extend the model with external tools, database connections, and custom data sources. The MCP configuration file is a JSON file that lists server definitions; Claude Desktop reads it at launch and makes the listed servers available as tools in the session.

# macOS config file location:
~/Library/Application Support/Claude/claude_desktop_config.json

# Windows config file location:
%APPDATA%\Claude\claude_desktop_config.json

Each MCP server entry specifies the server type, command, and any arguments. The same JSON format works on both macOS and Windows, which makes it straightforward to share a configuration across a team by committing the config file to a repository and having each developer copy it to the local path.

Installing the desktop clients

Download the appropriate installer from the release page linked from claude ai download. On macOS, mount the DMG and drag the app to Applications. On Windows, run the EXE installer with default settings. Neither installer requires Administrator rights on macOS; on Windows, a User-level install is available if you do not have admin access.

After the first launch, you will be prompted to sign in and connect an API key or account. The setup wizard walks through this in two or three steps. For teams using a shared workspace, the desktop client can also accept a workspace token rather than individual API keys, which keeps credential management central.

Research on human-AI interaction design — including work from Stanford HAI — is worth reading if you are evaluating which interface mode (CLI vs desktop) better supports your team's oversight requirements.